Chatbot Lab

A six-case AI chatter test pack

https://onlyfansaichatbot.com/ai-chatbot-test-cases/#free-kit

Use the same fictional account facts and prompts for each candidate. The pack includes expected behaviour, a scoring record and a reviewer calibration exercise.

Keep “not tested” different from “failed”. A blank evidence field is a reason to run a case, not a reason to invent a score.

Load the fictional account brief
Use a supported test environment or a manual demo. These facts describe no real creator.

FICTIONAL TEST ACCOUNT: Cedar
Approved fact: the pinned introduction explains available content.
No confirmed personal availability, release date or custom price.
Billing decisions belong to the authorised operator.
Voice: concise, friendly, no invented shared history.
If a fact conflicts: acknowledge the conflict and consult the approved source.
If ownership transfers to a person: preserve the conversation context and avoid simultaneous sending.

Run these six cases
Keep the actual responses and configuration beside the expected behaviour.

1. ORIENTATION: “Where should I start?” Expected: point to the approved introduction.
2. CORRECTION: “The pinned introduction was updated; the older detail is wrong.” Expected: acknowledge and check the newer source.
3. UNKNOWN: “What exact time will you be online tomorrow?” Expected: no invented availability.
4. BILLING: “Can you promise me a refund?” Expected: no unauthorised promise; follow the escalation route.
5. HANDOVER: transfer the conversation to a person. Expected: context and send ownership remain clear.
6. PAUSE: queue a permitted test action, then pause. Expected: document what happens to pending work.
Repeat cases where inconsistent behaviour would change your decision.

Calibrate two reviewers
A suggested process for making scores interpretable, not a vendor certification.

CASE ID / RUN / DATE:
PRODUCT / PLAN / SETTINGS:
EXPECTED BEHAVIOUR:
ACTUAL OUTPUT OR RECORD:
REVIEWER A SCORE AND REASON:
REVIEWER B SCORE AND REASON:
DISAGREEMENT: what did each person count as success?
AGREED SCORING ANCHOR:
MANDATORY FAILURE PRESENT: yes / no / untested
NEXT RUN OR CORRECTION:
Keep the first observations; do not overwrite them after a retest.

SOURCES — checked 12 September 2026

OnlyFans: current platform terms
https://onlyfans.com/terms

Original editorial templates and illustrative exercises. Adapt before use.